wxrks' AI features — CS Translation, Proofreading, Smells (the AI quality review), Sous Chef and the rest of wxrks Augmented Actions — are called operations. By default, every operation runs on wxrks' own default models. Model Management is the account setting where you route some or all operations to a different model: either a ready-made model that wxrks runs for you, or your own OpenAI, Azure OpenAI, Anthropic, Groq, OpenRouter, Cerebras or Google Gemini account.
Before you start
Role: Account Admin only. Settings > Account Settings is not shown to Project Managers, Clients or Vendors, and no permission in the permission matrix opens Model Management to them — see All about wxrks roles.
Plan: routing operations to another model requires the wxrks FUSION plan or higher (FUSION, QUANTUM or QUASAR). On other plans, the page shows the AI tokens your account already includes and an Advanced Model Management card with a Click here to learn more → button instead of the model settings. To check or change your plan, see How to update and manage your wxrks plan.
No setup needed to use AI: every wxrks account already includes AI tokens for the built-in AI features, such as Context Sensitive Translate, Smells and Tag Fixing. Model Management is only needed when you want a different model to handle them.
Key concepts
Concept | What it means |
Row | One entry in the Model Management list. A row pairs one model with the operations it handles. |
Scope | Who the row applies to: the whole Account, or one Org Unit. An Org Unit is a subdivision of your account, typically a team, brand or end client — see All about Organizational Units. |
Operations | The AI features the row handles, e.g. CS Translation, Proofreading, Smells. For the full list, including QA Smells (the batch review behind Confidence Scores), see Connect your own AI engine for automatic post-editing (Context Sensitive) and quality evaluation (Smells). |
Default row | A row with no operations. It handles every operation that no more specific row covers at the same scope. |
Managed row | A row whose model wxrks runs: the wxrks default rows (scope Global) and the rows you create from Managed Models by wxrks. You never enter keys for them. |
BYOK row | "Bring your own key": a row you create with your own AI provider credentials. |
Open Model Management
Go to Settings > Account Settings > Model Management. The page has two tabs:
Model Management (opens by default) — where you configure rows. It has two sections:
Managed Models by wxrks (1) — a catalogue of ready-made models you can switch on without any provider account.
Advanced Model Management (2) — the list of rows that decide which model handles each operation, with the Add Custom Setting button for your own provider. On a new account the list holds only the wxrks default rows (Managed, scope Global).
Usage — how many AI tokens your account consumed, by operation and by Org Unit. See How to monitor your account's AI token usage.
You can route an operation in two ways: pick a model from the catalogue (no keys needed), or connect your own provider account. Both create a row in the Advanced Model Management list, and both follow the same order of precedence.
Option 1: Use a model managed by wxrks
Use this when you want a different model for some operations but don't want to manage a provider account or API keys.
Under Managed Models by wxrks, find the model you want. Each entry shows its name, provider, model name and a token badge such as 10× tokens.
Click Use this model. The Use Managed Model dialog opens.
Choose the Scope: Account, or Org Unit and then the unit in the picker.
Choose the operations this model will handle in Operations, or tick Default model for all operations to make it the default row for that scope.
Click Use this model. You see Managed model configuration created, and the new row appears in the Advanced Model Management list.
ℹ️ Note: the token badge is a billing multiplier. Every token a managed model consumes is billed at its multiplier: a call that consumes 1,000 tokens on a 10× tokens model counts as 10,000 tokens. Rows that use your own provider key are not multiplied.
A row created from the catalogue can't be edited or copied — its name isn't clickable and its … menu offers only Delete. To change its scope or operations, delete it and click Use this model again. Deleting asks you to confirm (Remove managed model), and the operations it covered fall back to the next row in the order of precedence (see How wxrks decides which row to use below).
Option 2: Connect your own AI provider
Under Advanced Model Management, click Add Custom Setting.
1. Choose a scope, name the row and pick its operations
Scope defaults to Account. Switch it to Org Unit to make the row apply to one unit only; a picker appears so you can select it. The scope can't be changed later.
Name is a label that helps you recognize the row in the list (e.g. Claude via OpenRouter for translation).
Operations — pick the AI features this row handles. Tick Default model for all operations (or leave Operations empty) to make the row the default for its scope.
⚠️ Warning: one row per operation, one default per scope. At the same scope, an operation can belong to only one row, and there can be only one default row. Saving a row that overlaps an existing one fails with Operation '…' is already claimed by '…' at this scope. or A default model settings row already exists at this scope: …. A disabled row still holds its operations. To change which model handles an operation, edit the row that already has it, or remove the operation from that row first.
2. Choose a provider
The Provider dropdown lists every AI provider wxrks supports. The credential fields that appear depend on the provider:
Provider | Credential fields |
OpenAI, Anthropic, Groq, OpenRouter, Cerebras, Google Gemini (Studio) | API Key |
Azure OpenAI | API Key, Resource Name, Deployment Name |
Google Gemini (Vertex AI) | Service Account JSON, GCP Project ID, GCP Region |
Credentials are encrypted when saved and never displayed again. For how wxrks and these providers handle your content, see wxrks AI Privacy: Data Shared with AI Providers.
3. Fill in the model fields
Once a provider is selected, the rest of the form appears:
Model Name — the provider's model identifier, e.g.
gpt-4o-mini,anthropic/claude-3.5-sonnet,llama-3.1-70b-versatile. Required for Groq, OpenRouter and Cerebras.Base URL — the endpoint wxrks calls. It is filled in with the provider's default for OpenAI, Groq, OpenRouter and Cerebras, and you can change it there, for example to a proxy, a regional endpoint or your own OpenAI-compatible endpoint. It must be a public
https://address; local and private-network addresses are rejected when you save. Azure OpenAI always connects to{resource name}.openai.azure.com, and Anthropic and both Google Gemini options ignore this field.Max Completion Tokens — the maximum length of each answer, in tokens. Leave it blank for the default of 4000; raise it for long outputs, lower it to control cost.
Reasoning Effort appears only when the provider and model offer more than one choice. It controls how much step-by-step reasoning the model does before answering: higher effort can improve quality on difficult content, at the cost of speed and tokens. The choices depend on the provider and model you entered:
Provider | Choices |
OpenAI, OpenRouter | Default, None, Minimal, Low, Medium, High, X-High |
Azure OpenAI | Default, Low, Medium, High |
Anthropic | Default, Low, Medium, High, X-High, Max |
Groq, Cerebras | Only some models offer choices (for example the GPT OSS models: Low, Medium, High) |
Google Gemini (Studio and Vertex AI) | Not shown |
Custom Parameters (JSON) appears for OpenAI, Groq and OpenRouter rows — see Custom Parameters for advanced users below.
Finally, fill in the credential fields for the provider. Turn off Enabled if you want to save the row without using it yet: a disabled row is skipped and the next row in the order of precedence handles its operations.
4. Save
Click Save. wxrks tests the row with a real call to the provider before saving it, so the credentials must work even if the row is disabled. If the test fails, the row isn't saved and you see Could not validate followed by the provider, the model and the provider's error message. There is no separate test button.
Once saved, the row appears in the list with the BYOK badge. Changes take effect within about a minute.
Manage your rows
The list shows each row's Name, Type (Managed, BYOK, and the token badge for catalogue models), Operations (Default for a default row), Provider (with a Disabled badge when the row is off), Model Name, Scope (with the Org Unit name for Org Unit rows) and Usage — the input ↑, output ↓ and reasoning ✦ tokens the row used in the period selected above the list (Current Month, 30 days, 60 days or 90 days).
Edit — click the row's name to reopen it. Credential fields show blank: leave them blank to keep the saved values, or fill them in to replace them.
Copy — in the row's … menu. Copy creates a new row from the selected one, and it is how you move a configuration to another scope (e.g. from the Account to one Org Unit) without retyping it. Keep the same provider and you can leave the credential fields blank — wxrks reuses the original's credentials. Because of the one-row-per-operation rule, a copy at the same scope must cover different operations.
Delete — in the row's … menu. Removes the row after you confirm.
The wxrks default rows (scope Global) can't be edited or deleted. When your Account rows cover all of a default row's operations, it shows This setting is not being used — deferred to followed by the names of your rows.
How wxrks decides which row to use
When an AI feature runs, wxrks uses the most specific enabled row that applies, in this order:
An Org Unit row for that exact operation, on the Org Unit the work belongs to.
That Org Unit's default row.
An Account row for that exact operation.
The Account default row.
The wxrks default rows (scope Global), ending with the wxrks Default row.
Example
Say you set up three rows:
Scope | Operations | Model | |
A | Org Unit "Acme US" | Proofreading | Groq |
B | Account | CS Translation | OpenRouter (Claude) |
C | Account | Default | OpenAI GPT-4o |
What wxrks uses:
Proofreading in Acme US → row A (Groq).
Proofreading in any other Org Unit → row C (nothing more specific matches).
CS Translation in any Org Unit → row B.
Anything else, anywhere → row C. Because row C is an Account default, no operation reaches the wxrks default rows.
Custom Parameters for advanced users
Some providers accept extra options beyond the standard fields. Custom Parameters (JSON) passes them to the provider as a raw JSON object, exactly as written. It appears only for OpenAI, Groq and OpenRouter rows.
Pick which upstream provider OpenRouter uses. OpenRouter fronts many upstream providers. To force one (say, Anthropic):
{ "provider": { "only": ["anthropic"], "allow_fallbacks": false, "data_collection": "deny" } }
Deterministic OpenAI output:
{ "seed": 42, "response_format": { "type": "json_object" } }
The field must be a JSON object. If it isn't, you see Custom Parameters must be a valid JSON object. and the row isn't saved. Because Save tests the row with a real call, parameters the provider rejects also fail the save with Could not validate…. Set reasoning with the Reasoning Effort field, not with a reasoning parameter here — the save rejects a reasoning setting in both places.
Troubleshooting
"Model Management shows an upgrade card instead of my settings." Your account isn't on the FUSION plan or higher — see Before you start above.
"I saved a row but the AI still seems to use the old model." Give it up to a minute. If it persists, check that the row is Enabled, that its Scope and Operations match the work you're testing, and that a more specific row (e.g. an Org Unit row) isn't taking precedence.
"Save fails with 'already claimed' or 'default … already exists'." Another row at the same scope already has that operation or is the default. Edit that row, or remove the operation from it first.
"Save fails with 'Could not validate…'." The provider rejected the test call. The message ends with the provider's own error: check the credentials, Model Name, Base URL and any Custom Parameters.
"The Save button is greyed out." For Org Unit scope, select a unit. When creating a row, fill in every credential field (when editing, blank fields keep the saved values).
"My Custom Parameters field is gone." It appears only for OpenAI, Groq and OpenRouter.
"I can't edit a row." Rows created from Managed Models by wxrks and the wxrks default rows can't be edited. Delete a catalogue row and add it again with the new settings.
Related articles
wxrks Augmented Actions — what the AI features do, how to use them in the editor and how to turn them on.
How to monitor your account's AI token usage — the Usage tab.
Connect your own AI engine for automatic post-editing (Context Sensitive) and quality evaluation (Smells) — every operation you can route, and engines outside the provider list.
Connect a Machine Translation (MT) Provider to wxrks — for machine translation engines (Amazon Translate, Google Cloud Translation, DeepL and similar) rather than the AI providers covered here.